Fix to_pyg graph attribute mutation

[pragnyanramtha]

Summary

• copy the input graph before pruning node attributes for PyG export
• keep exported PyG graphs limited to requested node features
• add a regression test proving the caller graph remains unchanged

Fixes #139.

Validation

• PYTHONPATH=src python -m pytest tests/test_save.py::test_to_pyg_does_not_mutate_node_attributes -q
• PYTHONPATH=src python -m pytest tests/test_save.py -q
• python -m black --check src/weather_model_graphs/save.py tests/test_save.py
• python -m isort --check-only src/weather_model_graphs/save.py tests/test_save.py
• python -m flake8 src/weather_model_graphs/save.py tests/test_save.py
• git diff --check

[Copilot]

Pull request overview

Fixes weather_model_graphs.save.to_pyg() mutating the caller’s NetworkX graph by pruning node attributes on a copied graph prior to PyTorch Geometric conversion, and adds a regression test to prevent reintroduction.

Changes:

• Copy the input graph inside to_pyg() and prune node attributes on the copy (not the caller-owned graph).
• Ensure downstream conversion/splitting logic uses the copied/pruned graph consistently.
• Add a regression test that asserts the input graph’s node attributes remain unchanged while the exported/converted graph excludes unrequested attributes.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
src/weather_model_graphs/save.py	Avoids mutating the input graph by copying before pruning node attributes and exporting.
tests/test_save.py	Adds a regression test verifying to_pyg() does not delete node attributes on the caller graph.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.